Ratherthan framing AMP therapy as “personalized,” a more clinically grounded strategy involvespathogen-directed or susceptibility-guided AMP design informed by resistance phenotypesand epidemiological data

unctional annotations of unknown sequences, genetic associations,disease links, and evolutionary relationships, their capacity to directly predict antimicrobialefficacy remains under active development

Pareto front (PF), by the “filtration” mechanisms toidentify the “non-dominated” AMPs made by the Pareto ranking (PR)

wherein EC50 refers to the concentration producing 50% ofthe maximal biological effect,

Most AI models achieve high accuracy on predictinghemolytic and non-hemolytic AMPs; however, they also have a limitation on predictinghigh vs low hemolytic AMPs

Thus, thechallenges mentioned emphasize the necessity of new delivery systems and formulationstrategies

many AI tools do not cover the five levels of AMP discovery and design [

MPGen uses an order-agnostic autoregressive diffusion model

which hasa limit of 30 amino acids for AMPs prediction sequences.

Antimicrobial Peptide Database (ADP), the Collection of Anti-MicrobialPeptides (CAMP) and the Database of Anuran Defense Peptides (DADP)

oreover, substitution with DAAs may enhance other properties, such asantimicrobial activity, by increasing permeability into bacterial membranes and eukaryotictissues

macrocyclization, which can be achieved by the con-densation of their amino and carboxyl group terminals

However, a concern in the incorporation of liposomes in cationic AMPs has emerged, asleakage of these antimicrobials can occur due to hydrophobic and electrostatic interac-tions [210, 211 ]. The high production cost of liposomes

s molecular weight increases, the half-lives, retention in thebody and resistance to protease degradation increase, requiring a more complex metabolicprocess in the human body to excrete

Furthermore, numerous AFPs not only eradicate fungi solely by membranelysis but also cause oxidative stress, mitochondrial dysfunction, programmed cell deathand autophagy

fungal membranes contain ergosterol (the main fungalsterol) and glycosphingolipids, more specifically glucosylceramide (GlcCer), which areboth surrounded by a solid cell wall composed of chitin and β-glucans

bacterial membranes consist of a high content of anionic phospholipidand strong electrostatic attraction to cationic AMP

ifferences in AMP selectivity between fungal and bacterial cells toprecise AMPs and AFPs have to do with the amount of ergosterol levels,

From a biochemical perspective, the plasma membrane controlsfluidity, permeability and possesses the ability to respond to stress

bioactive peptides, which are derived from different natural sources involving in-sects, plants and marine organism

They aredefined as catatonic molecules with an affinity for cell membranes, where the activity iscontrolled by distinct properties of the cell

C. auris producesextracellular polymeric substances that promote surface adhesion, biofilm formation andresistance to antifungal agents. These properties facilitate persistence on medical devices

fungaldiseases affect billions of people annually and are associated with several million deathsworldwide,

including combinations with chitin, cholesterol, sodium alginate andespecially chitosan nanoparticles

the mechanisms of virulence in Enterobacter spp. have yet tobe studied thoroughly

there is protein synthesis,phospholipases, capsular polysaccharides, lipopolysaccharides, membrane and QS pro-teins, metal absorption, cytokines and biofilms [ 152 ,156 ].

In the case of biofilms, QS is important for biofilm formation andcommunication or coordination within the biofilm

Another mechanism of resistance against AMPs is biofilm formation, which is acommon virulence factor from ESKAPE pathogens

which makes it more difficult for pathogens to develop resistance, as it lowersthe mutation rate.

binding to membrane componentsto disrupt membrane stability or synthesis and penetration of the cell to alter vital cell pro-cesses that ultimately lead to cell death

facilitating the design of peptides that selectively promote aggregation of bacterialproteins while minimizing off-target self-assembly or host toxicity.

amyloidogenic and aggregation-prone properties can also be ex-ploited as a functional antimicrobial mechanism

AI-guided aggregationand solubility models play a central role in de-risking AMP development

support the design of protectivemodifications such as D-amino acid substitution, cyclization, PEGylation, terminal cap-ping, or incorporation of noncanonical residues

analyze charge distribution, hydrophobic moment, amphipathicity, sequencepatterns and predicted secondary structure to identify features associated with erythrocytelysis

hemolyticeffects, rapid degradation by proteases, poor solubility, or aggregation tendencies

Such models help distin-guish peptides that preferentially target microbial membranes from those that may damagemammalian cells

ackbone sampling, side-chainoptimization and energy-based refinement

AlphaFold, powered by DL and attention-based architectures

Figure 2

The agent then updates its policy to favor sequence patterns associated with higherrewards

reverse diffusion process

peptide design, enabling controlled modifications and multi-objective op-timization

a generator produces peptide sequences and a discriminator evaluates their resemblance toknown AMPs

generative models are capable of creatingentirely new peptide sequence

controlled sampling, latentspace manipulation and iterative optimization,

antimicrobialactivity emerges not only from amino acid composition but also from structural elementssuch as secondary structure, amphipathicity, solvent accessibility and conformational stabil-ity

apture long-range contextualrelationships through self-attention mechanisms

hydrophobic patches, cationic regions, or amphipathic periodicity that contributeto membrane interaction

nonlinear relationships, capturelong-range dependencies

SVM, RF and XGBoos

nonlinear interactions between features

n AMP discovery pipelines, RF models frequently outperform single-tree methods byreducing variance and improving generalization across diverse peptide families

ecision trees trained on bootstrapped subsets of the data

esilience to overfitting

yperplanes in high-dimensional feature space

osition-specific scoring matrices,

“handcrafted descriptors”

k-mer frequencies, dipeptideand tripeptide motifs, pseudo-amino acid composition

et charge,hydrophobicity indices, hydrophobic moment, Boman index, aliphatic index, instabilityindex and predicted isoelectric point

Featureextraction,

These methods rely on the systematic transformationof peptide sequences into informative numerical representations

When immobilized onbiomaterials, AMPs can prevent bacterial colonization, disrupt early biofilm formationand extend the functional lifespan of medical device

de novo peptide design and multi-objective optimization, integrating activity, selectivityand stability considerations into the design process

toxicity toward mammalian cells, poor serum stability, limited bioavailability

such as membrane disruption through pore formation